For example, a company may want to detect occurrences of their account numbers, which are 10-digit numbers. However, there are a lot of 10-digit numbers in the world, and such a broad detection rule could be very noisy. One option is to add context to the rule, such as the appearance of the strings “account” or “acct” near the token. An appearance of the token along with the context will be upweighted to a “very likely” violation, whereas the token without the context would be weighted as “possible.”